The Environmental Kuznet Curve for Deforestation

CERDI, Etudes et Documents, E 2012.16

C E N T R E D ' E T U D E S

E T D E R E C H E R C H E S

S U R L E D E V E L O P P E M E N T

I N T E R N A T I O N A L

Document de travail de la srie

Etudes et Documents

E 2012.16

The environmental Kuznets curve for deforestation:

a threatened theory? A meta-analysis

Johanna CHOUMERT, Pascale COMBES MOTEL,

Herv K. DAKPO

avril 2012

CERDI

65 BD. F. MITTERRAND

63000 CLERMONT FERRAND - FRANCE

TEL. 04 73 17 74 00

FAX 04 73 17 74 28

www.cerdi.org


Les auteurs

Johanna Choumert

Associate Professor, Clermont Universit, Universit dAuvergne, CNRS, UMR 6587, Centre dEtudes et de

Recherches sur le Dveloppement International (CERDI), F-63009 Clermont-Ferrand, France

Email : [email protected]

Pascale Combes Motel

Professor, Clermont Universit, Universit d'Auvergne, CNRS, UMR 6587, Centre dEtudes et de

Recherches sur le Dveloppement International (CERDI), F-63009 Clermont-Ferrand, France


Herv K. Dakpo

Student, Clermont Universit, Universit d'Auvergne, CNRS, UMR 6587, Centre dEtudes et de Recherches

sur le Dveloppement International (CERDI), F-63009 Clermont-Ferrand, France


Corresponding author: [email protected]

La srie des Etudes et Documents du CERDI est consultable sur le site :

http://www.cerdi.org/ed

Directeur de la publication : Patrick Plane

Directeur de la rdaction : Catherine Araujo Bonjean

Responsable ddition : Annie Cohade

ISSN : 2114-7957

Avertissement :

Les commentaires et analyses dvelopps nengagent que leurs auteurs qui restent seuls

responsables des erreurs et insuffisances.


Highlights We carry out a meta-analysis of Environmental Kuznets Curve (EKC) studies for

deforestation. We analyze the results of 71 studies, offering a total of 631 estimations. We investigate the incidence of choices made by authors (econometric strategy,

deforestation measure, temporal coverage, geographical area, measure of economic development) on the probability of finding an EKC.

After a phase of work corroborating the EKC, we find a turning point after the year 2001.

Abstract

Although widely studied, deforestation remains a topical and typical issue. The relationship

between economic development and deforestation is still at stake. This paper presents a meta-

analysis of Environmental Kuznets Curve (EKC) studies for deforestation. Using 71 studies,

offering 631 estimations, we shed light on why EKC results differ. We investigate the incidence

of choices made by authors (econometric strategy, deforestation measure, temporal coverage,

geographical area, measure of economic development) on the probability of finding an EKC.

After a phase of work corroborating the EKC, we find a turning point after the year 2001.

Building on our results, we conclude that the EKC story will not fade until theoretical

alternatives will be provided.

Keywords

Meta-Analysis; Environmental Kuznets Curve; Deforestation, Development

JEL Codes

C12, C83, Q23, O13


1

Mark Twain: The report of my death was an exaggeration

New York Journal 2 June 1897

1. Introduction

One important question which is still at stake, even more in an economic crisis

context, is whether environmental and economic objectives are compatible subjects.

This question gets high resonance in the literature devoted to the Environmental

Kuznets Curve (EKC) for deforestation which is a subject of confrontation between

optimistic and pessimistic views of development (Carson 2010). This literature has

gained considerable expansion in economics as well as in natural sciences (Mather et al.

1999). From the 1990s onwards, numerous studies following the idea of (Grossman &

Krueger 1995) tested an EKC for deforestation.

Studying EKCs for deforestation is an important matter for two reasons.

First, forest depletion has been given increasing attention especially with the

recognition of the role of forests in the global carbon cycle. Recent developments in the

literature on climate change put forward the potential role of forests in climate change

mitigation (N. H. Stern 2007). The impetus was given to the so-called Reducing

emissions from deforestation and forest degradation (REDD) device first proposed for

negotiation at the Montreal UNFCCC CoP held in 2005. The Bali Road Map adopted in

2007 went beyond with the REDD+ concept by enhancing other objectives such as

sustainable management of forests and valuation of forests carbon stocks (Angelsen,

Brockhaus, et al. 2009; Angelsen, Brown, et al. 2009; Engel & Palmer 2008). As pointed

out by (Geist & Lambin 2001) the role of forests in mitigating global environmental

threats such as climate change and biodiversity erosion is a research imperative and has

been motivating considerable efforts to the understanding of patterns and causes of the


2

deforestation process and in fine to the derivation of policy implications (Angelsen &

Kaimowitz 1999).

Second, EKC has drawn attention from scholars for several years (e.g. (Ma & D. I.

Stern 2006). Studying EKCs for deforestation may therefore offer an interesting case in

the methodology of economics from a sociological point of view. According to (Kuhn

1996), scientists share common beliefs and practices which circumscribe the normal

science. They are often reluctant to diverge from them since they are most of the time

evaluated and published according to rules widely accepted within the normal science.

EKCs seem to belong to such a set of practices. It is indeed rather striking that several

researchers seem to consider it as econometrics as usual i.e. a relevant hypothesis

which helps understanding environmental degradation and for instance the

deforestation process. Other researchers have conducted critical reviews and forcefully

argued in favour of its obsolescence (Levinson 2002; D. I. Stern 2004). Thus there seems

to be a discrepancy between, on the one side, researchers who dismiss EKCs and, on the

other side, scientists who consider that EKCs are relevant. Among the latter, the EKC is

still presented as one of the hypotheses explaining the forest transition process (Barbier

et al. 2010; Rudel et al. 2005; Mather 1992).

The objective of this paper is to examine why scholars get accustomed to the EKC for

deforestation or whether they have objective reasons to dismiss it. In contrast with

existing studies or critical surveys on the EKC for deforestation, we propose to address

the subject with a meta-analysis which can be considered as an attempt to identify

wheat from chaff (Stanley 2001) i.e. identify a set of reasons that favour or falsify

empirical evidences of EKCs for deforestation.


3

The rest of the paper is organised as follows. Section 2 presents the debate in the

literature on the EKC for deforestation. Section 3 further justifies the use of meta-

analysis. Section 4 is devoted to the construction of the database. Section 5 develops on

the empirical strategy and the econometric results. Section 6 discusses the results and

Section 7 concludes.

2. Contrasted Premises and Results

The story with EKCs started with (Grossman & Krueger 1995) and (Panayotou 1993).

According to (Beckerman 1992, p.482) : [] there is clear evidence that, although

economic growth usually leads to environmental deterioration in the early stages of the

process, in the end the best - and probably the only - way to attain a decent environment

in most countries is to become rich. This optimistic premise was later mitigated by the

(Arrow et al. 1996) assertion according to which economic growth is not a sufficient

condition for getting environmental improvements or curbing environmental

degradation. It is even more doubtful to find EKCs when stock variables with strong

potential feedback are expected in, for instance, ecosystems such as tropical forests.

The economic literature provides theoretical underpinnings for an EKC. Many of them

model pollutants emissions income relationships (Andreoni & Levinson 2001;

Xepapadeas 2005). (Munasinghe 1999) develops a static model showing that

appropriate policy measures (i.e. removal of subsidies on timber exports) could help

alleviating the pressure on tropical forest stocks and thus tunnelling through the EKC

for deforestation. Theoretical explanations of EKCs for deforestation were also

developed with the so-called poverty-environment hypothesis as opposed to the capital

or frontier model (Geist & Lambin 2001; Rudel & Roper 1997). The contention that

economic development has a negative (positive) impact on deforestation is suggested by


4

the former (latter). Theoretical models explaining agricultural land expansion may also

be compatible with an EKC for deforestation (Barbier 2001). It is however worth to

notice that several authors are rather unsatisfied with the current theoretical literature

on the EKC for deforestation and call for further developments (Roy Chowdhury &

Moran 2012).

Figure 1. Empirical analyses of EKCs for deforestation

Source: Own calculations. See Section 4 for more details on the construction of the database. One observation represents one estimation.

Early econometric results (Shafik & Bandyopadhyay 1992) and further ones on the

existence of an EKC for deforestation were not conclusive. More than a half over more

than 600 estimations published since 1992, do not corroborate the EKC (Figure 1). This

bulk of mixed results seems to have encouraged several authors against the EKC. One of

the most serious attack against EKCs was provided by (D. I. Stern 2004; Levinson 2002)

with close titles. Critical reviews of EKC focusing on deforestation raised several

empirical shortcomings which may have impacted the EKC for deforestation

0

20

40

60

80

100

120

Nu

mb

er

of

ob

serv

ati

on

s

Year of publication

EKC No EKC


5

corroboration. As mentioned by (Koop & Tole 1999), FAO databases were mostly used.

Early FAO databases were from FAO production yearbook. Supposedly, more recent

ones from Forest Resources Assessment published in most recent studies have greater

reliability. Irrelevant econometric methods are also put forward. Early estimates relied

on cross-section data which imply restrictive hypotheses (Dinda 2004; Koop & Tole

1999). More recent studies thus implemented panel data or more generally pooled cross

sectional time series data. It may be worth to examine whether improvements made in

econometric devices had an impact on the existence of an EKC for deforestation.

One may thus be confused between on the one side, Stern's contention according to

which most of the EKC literature is econometrically weak (D. I. Stern 2004, p.1420)

and empirical analyses still relying on or testing the EKC for deforestation. But one may

understand the situation while remembering (McCloskey 1983). It seems that

economists need more than statistical tests to get rid of a hypothesis like the EKC. Would

they give weight to the rhetoric surrounding the EKC? In this paper, we want to check

whether characteristics of publications dealing with EKCs for deforestation play a role in

their corroboration. We argue that conducting a meta-analysis of EKCs for deforestation

equations can contribute to their critical assessment.

3. The Use of Meta-Analysis

A meta-analysis produces, through a set of statistical techniques, an overall summary

of empirical knowledge on a specific topic. Meta-analysis is the analysis of empirical

analyses that attempts to integrate and explain the literature about some important

parameter (Stanley & Jarrell 1989, in (Stanley et al. 2008, p.163). It can explain the

excess study-to-study variation typically found in empirical economics, uncover the

trace of statistical power that is associated with a false theory, and see through the


6

distortion of publication selection and misspecification biases (Stanley 2005, in

(Stanley et al. 2008, p.163).

Before the 90's, scientists synthesised the results of the research literature through

the use of narrative review. However, the limit of this method was quickly reached with

the increase in the amount of information available. Indeed it seems inconceivable using

this method to successfully synthesize the results of dozens of studies which, moreover,

are conflicting in their results. (Hunter & Schmidt 2004) explain that narrative surveys

developed the so-called myth of the perfect study: a researcher may be convinced that

some studies suffer from methodological flaws and exclude them from the review;

another one may consider different methodological flaws. Hence, idiosyncratic ideas

may lead one to reduce available literature on a topic from 100 studies to 10 and

another one from 100 to 50.

Meta-analysis allows to address the inherent limitations of the traditional narrative

review by providing a more effective approach (Stanley 2001); it allows to set up an

evaluation with several measures of outcome; it permits the use of moderators: some

characteristics of the samples used in different studies influence the results obtained.

Moreover, meta-analysis is a good tool for decision support because it helps in the

assistance in planning, in the generation of new hypotheses, and eventually, prospective

meta-analysis.

Research questions are often the subject of numerous studies leading to various

controversies. As mentioned earlier, the EKC for deforestation is no exception, on the

contrary, as shown in different reviews of the literature on the subject. Different

opinions differ on the question of the existence of an EKC for deforestation and this,

especially since empirical analysis validate and invalidate this hypothesis according to


7

econometric models, measures of deforestation, study area, temporal coverage, etc.

Since 1992, empirical work has been accumulated and we even see a significant increase

in the number of studies since the 2000s (Figure 1). Finally, despite the abundance of

empirical studies, it is difficult to propose clear recommendations arising directly from

the relationship between economic growth and deforestation. Our approach, based on

the meta-analysis, is complementary with respect to existing surveys (D. I. Stern et al.

1996; D. I. Stern 2004; Dinda 2004). Indeed, it allows us to statistically analyse the

variation in results found between existing works and in particular to analyse the

influence of choices made by the authors on their results (Stanley 2001).

There exists meta-analysis for EKC studies. These encompass however various

environmental goods. (Cavlovic et al. 2000) performed a meta-analysis of 25 studies -

121 observations - with a focus on turning points. They cover 11 categories of

environmental goods such as urban air quality or deforestation. Observations for

deforestation represent only 6 % of their sample. The impact of deforestation on income

turning points depends on the econometric specification of the meta-analysis. (Jordan

2010) analyses 255 articles - 373 observations - among which 8 % deal with

deforestation. Results show no significant impact of deforestation on the probability of

finding an EKC.

The operationalization of a meta-analysis is done by following specific steps. The first

step is the formulation of the subject: what are the issues and objectives of the study

(central questions we would like answers to) and what are the assumptions made. The

second step is the overall design of the study: this is to clarify outcomes, populations,

settings, criteria for inclusion and exclusion of studies. The third step is to specify the

sampling plan, the units of observation and then begin the literature search. The fourth


8

step is the extraction of data: it is directly applicable to studies that were selected for

further analysis after the application of inclusion / exclusion criteria. The fifth step is the

statistical analysis. As recommended by (Nelson & Kennedy 2008), we address relevant

issues such as heteroskedasticity, the non-independence of regressions, the specification

of the model, the non-normality of residuals and multicollinearity.

4. Construction of the Database

4.1. Sampling procedure and sample characteristics

This meta-analysis is based on 71 studies, offering 631 estimations.1 In a meta-

analysis, an important aspect of the work is to find studies that addressed the theme of

the investigation. We conducted our research using academic databases such as Science

Direct, Springer, RePEc Ideas, Mendeley, Wiley, etc. The first step of this research

was to enter keywords such as Kuznets AND Deforestation, Income AND

Deforestation, Development AND Deforestation, Poverty AND Deforestation. Given

that there are relevant studies which are not identified by bibliographic databases, in a

second step we carried out a manual search and this by identifying journals and articles

relevant to the topic. We then browsed these studies in order to identify new ones which

have addressed closely the subject. A third step was necessary as we had to directly

contact the authors since some studies are not accessible on the web. At the end of the

research process, we got to build up an information base of nearly 102 items that we

have then filtered through inclusion criteria.

At this stage, we selected eligibility criteria to be applied to the bibliographic

database. In order not to bias the meta-analysis on subjective considerations, we tried to

be broad in our inclusion criteria. First, we looked closely at the dependent variable

1 The list of primary studies is available from the authors upon request (See Table A - 1 in the Appendix).


9

used in each estimation. Once done, we decided to retain all the measures relating

directly to deforestation. At the end of this inclusion work, we have reached 71 studies

that were used to build the database of this article.2

Every relevant estimation was carefully examined and the information was gathered

to build the database. Accordingly, we surveyed all the information given by the authors

related to the EKC (coefficients associated with variables of interest), the sample (size,

geographical area ), the econometric method, control variables, and all information

considered necessary to carry out this meta-analysis. In terms of coefficients and their

sign, they were adjusted according to the nature of the dependent variable and

depending on the outcome. For example, when the dependent variable is the rate of

deforestation, to obtain an EKC, the coefficient of per capita GDP should be positive and

negative for the squared GDP. By cons, if the dependent variable is the stock of forest,

the hypothesis of an EKC is confirmed if the coefficient of GDP is negative and that of

squared GDP is positive. Also to be as broad as possible we defined our null hypothesis

whenever the estimated marginal effects of GDP and GDP squared had the expected

signs at the 10% level.3 We did not try to minimise this type I error since it is

detrimental to the power of the test. In brief, we define our dependent variable as EKC

which equals to 1 when the existence of the EKC for deforestation is corroborated and 0

otherwise.

2 For instance were excluded from the sample studies analyzing biodiversity, the ecological footprint, land vulnerability or protected areas. Also studies focusing on households were excluded.

3 The decision on the results was based on the significance of all coefficients of interest. For instance, in the quadratic model, as in most studies tests for joint significance were not performed, the model is considered significant if the two coefficients (i.e. per capita GDP and squared per capita GDP) are individually significant.


10

4.2. Moderator variables

At the end of this work, the database we had raw on hand included 631 observations

and numerous differences, most of which focused on the explanatory variables used.

Therefore for the sake of synthesis, these variables were grouped into classes. Table 1

presents the description and descriptive statistics of the variables. Next we shall discuss

factors that may affect the income-deforestation relationship. These are the explanatory

variables used in our model.

In view of the criticisms of the EKC literature, our variables of interest are the year of

publication and the econometric technique.

Regarding the time period, variables are the year of publication of the study, the

initial year of data, the final year of data. Differences between studies may reflect

numerous issues such an academic trend, the increasing quality and availability of

deforestation data, the mix of old and recent data or econometric requirements in peer-

reviewed journals. For instance, (Jordan 2010) finds that the most recent year of

publication is, the more likely to find an EKC decreases, while controlling for the nature

of environmental degradation.

With respect to econometric techniques, as pointed out in critical reviews of EKCs,

one may cast doubts on econometric techniques in early works. Advances in

econometrics may make recent studies more credible than they were in the nineties.

Several authors argued for instance that panel data regressions are more appropriate

for analysing the EKC. Indeed, deforestation is a dynamic process therefore the nature of

the relationship between income and deforestation should also be dynamic. If the time

dimension is ignored, a spurious income-deforestation relationship could be observed.

(Dinda 2004) argues that cross-section analysis is based on restrictive econometric


11

assumptions as to admit the same constant. What is more, using panel data also allows

for controlling for substantial time-invariant unobservable variables (such as countries

physical characteristics or historical factors among many others). On top of that, the EKC

theory was developed in an a-spatial framework. Hence, we could expect the inclusion of

spatial dimension to alter the EKC hypothesis.

In addition to the variables of interest, we propose control variables or moderators.

Concerning the measures of deforestation, we note that in early studies, the latter is not

homogeneous. As part of this work 15 different measures are used. It is interesting to

investigate whether these modify the income-deforestation relationship. In fact, some

measures of deforestation are causes of the latter, as the expansion of agricultural land

or infrastructures. (Geist & Lambin 2001) illustrate the differentiated effects of different

causes of deforestation. We also believe relevant to include the geographical area; in fact

growth path are different in developed and developing countries which can influence

the occurrence of EKC and even more so that the samples of studies are based on recent

data that do not capture development phases of developed countries. As for the

measures of economic development, we observe some heterogeneity in the sample.

However, there is no a priori on why and in what way such differences affect the

occurrence of an EKC. Regarding the publication type, according to the aim of the study,

one can expect different strategies of the authors, without compromising their

neutrality. In particular, as Lindhjem (2007) finds a significant effect of masters thesis

in his meta-analysis, we can also expect an effect of this publication type. In our case, we

suspect that Master students have less incentive to find an EKC, ceteris paribus. We also

include the presence of control variables for inequalities, life expectancy4 and

4 Life expectancy is correlated with environmental quality (Mariani et al. 2010).


12

infrastructures. We expect the inclusion of these variables to alter the income-

deforestation relationship. In fact, they are correlated with income and this may rise

collinearity issues or shed light on transmission channels. Finally we add other sources

of variation namely the sample size, the h-index of the publication,5,6 the type of model

estimated (quadratic or cubic) and the number of regressions presented.7,8

Table 1. Description and descriptive statistics of variables

Variables Description Mean SD Min Max

EKC = 1 if an EKC is found, 0 otherwise 0.330 0.470

Year Publication year 2004.582 4.705 1992 2011

initial data Initial year of data 1978.182 12.908 1948 2006

end data Final year of data 1997.403 5.475 1984 2010

sample size Sample size 73.643 211.526 4 2619

H index H index of the publication 0.518 0.412 0.000286 1.000286

reg Number of regressions presented in the study 24.324 19.81 1 60

Measures of deforestation

deforestation1 = 1 if Agricultural land expansion rate, 0 otherwise 0.019 0.137

deforestation2 = 1 if Arable land area, 0 otherwise 0.016 0.125

deforestation3 = 1 if Built-up land, 0 otherwise 0.005 0.069

deforestation4 = 1 if Crop Area, 0 otherwise 0.040 0.195

deforestation5 = 1 if Deforestation, 0 otherwise 0.777 0.417

deforestation6 = 1 if Forest area clearcut, 0 otherwise 0.013 0.112

deforestation7 = 1 if Forest stock, 0 otherwise 0.071 0.258

deforestation8 = 1 if Land use for fuel wood, charcoal, paper, and pulp, 0 otherwise

0.005 0.069

deforestation9 = 1 if Land use for pasture, 0 otherwise 0.005 0.069

deforestation10 = 1 if Land use to produce timber, 0 otherwise 0.005 0.069

deforestation11 = 1 if Mangrove deforestation index, 0 otherwise 0.006 0.079

deforestation12 = 1 if Paved roads in forest regions, 0 otherwise 0.029 0.167

deforestation13 = 1 if Roads encroaching on forest cover, 0 otherwise 0.006 0.079

5 The lower, the better

6 This ranking is done by Ideas Repec. At the date of last visit, journals were ranked from 1 to 3 496. So we calculated a ratio that is equal to the rank of the journal divided by the rank of the last journal listed (i.e. 3 496). However for studies that did not receive a ranking on this site, the ratio calculation was made assuming that the rank of this study will be equal to the rank of the last journal classified to which was added one.

7 Note that few studies provide tests to compare models. It is therefore not possible to provide a single regression per article.

8 Other information was collected from studies but was excluded due to collinearity issues such as the unit for the deforestation measure, the source of data for deforestation, etc.


13

Variables Description Mean SD

deforestation14 = 1 if Timber harvest rate, 0 otherwise 0.002 0.040

deforestation15 = 1 if Urban forest canopy percentage, 0 otherwise 0.003 0.056

Econometric techniques

econometrics1 = 1 if Dynamic Panel, 0 otherwise 0.016 0.125

econometrics2 = 1 if Geographically Weighted Regression, 0 otherwise 0.003 0.056

econometrics3 = 1 if Least Absolute Deviation, 0 otherwise 0.002 0.040

econometrics4 = 1 if Logistic Panel Smooth Transition Regression, 0 otherwise 0.016 0.125

econometrics5 = 1 if Monte Carlo Estimation, 0 otherwise 0.006 0.079

econometrics6 = 1 if Ordinary Least Squares, 0 otherwise 0.434 0.496

econometrics7 = 1 if Panel Fixed Effects, 0 otherwise 0.387 0.487

econometrics8 = 1 if Panel Random Coefficients, 0 otherwise 0.013 0.112

econometrics9 = 1 if Panel Random Effects, 0 otherwise 0.082 0.275

econometrics10 = 1 if Spatial Panel Fixed Effects, 0 otherwise 0.019 0.137

econometrics11 = 1 if Spatial Panel Random Effects, 0 otherwise 0.013 0.112

econometrics12 = 1 if Time Series, 0 otherwise 0.010 0.097

fe_retain = 1 if Panel Fixed Effects chosen with Hausman test, 0 otherwise 0.371 0.483

re_retain = 1 if Panel Random Effects chosen with Hausman test, 0 otherwise 0.043 0.203

Geographical areas

area1 = 1 if Africa, 0 otherwise 0.043 0.203

area2 = 1 if Africa and Asia, 0 otherwise 0.014 0.119

area3 = 1 if Africa, Asia, North America, Latin America and Oceania, 0 otherwise 0.019 0.137

area4 = 1 if Africa, Asia, North America and Oceania, 0 otherwise 0.006 0.079

area5 = 1 if Africa, Asia and Latin America, 0 otherwise 0.301 0.459

area6 = 1 if Africa, Asia, Latin America and Europe, 0 otherwise 0.008 0.089

area7 = 1 if Africa, Asia, Latin America and Oceania, 0 otherwise 0.006 0.079

area8 = 1 if Asia, 0 otherwise 0.086 0.280

area9 = 1 if Asia and Latin America, 0 otherwise 0.070 0.255

area10 = 1 if Europe, 0 otherwise 0.024 0.152

area11 = 1 if Latin America, 0 otherwise 0.073 0.260

area12 = 1 if North America, 0 otherwise 0.022 0.147

area13 = 1 if World, 0 otherwise 0.328 0.470

Economic development measures

development1 = 1 if GDP, 0 otherwise 0.642 0.480

development2 = 1 if GNP, 0 otherwise 0.068 0.252

development3 = 1 if Household median income, 0 otherwise 0.003 0.056

development4 = 1 if Human Development Index, 0 otherwise 0.165 0.371

development5 = 1 if Per capita consumption, 0 otherwise 0.011 0.105

development6 = 1 if Poverty, 0 otherwise 0.010 0.097

development7 = 1 if Urbanization, 0 otherwise 0.101 0.302

Publication types

pub type1 = 1 if Article, 0 otherwise 0.498 0.500

pub type2 = 1 if Conference Paper, 0 otherwise 0.024 0.152

pub type3 = 1 if Discussion Paper, 0 otherwise 0.013 0.112

pub type4 = 1 if Other, 0 otherwise 0.166 0.373


14

pub type5 = 1 if Working Papers, 0 otherwise 0.300 0.458

Presence of control variables

life expectancy = 1 if life expectancy variable, 0 otherwise 0.002 0.040

inequality = 1 if inequality variable, 0 otherwise 0.024 0.152

infrastructures = 1 if infrastructures variable, 0 otherwise 0.063 0.244

Others

cubic = 1 if Cubic model, 0 otherwise .0443 0.206

quadratic = 1 if Quadratic model, 0 otherwise 0.956 0.206

N = 631

5. Empirical Strategy and Results

Our response variable is a binary variable "EKC".9 Recall that we investigate the

factors that influence the occurrence of an EKC.10 We first estimate the model using

robust ordinary least squares.11 Indeed, in the presence of a binary dependent variable,

if the model is estimated by OLS, the errors are heteroskedastic. Therefore, it is

appropriate to make a robust estimation with White-corrected standard errors.

Furthermore, given that there may be a within-study autocorrelation leading to the

dependence of regressions within one article, we ran OLS with cluster-robust

inference.12 The reason is that each study estimates various regressions with different

units of measures (cf. Table 2). Then, we perform a bootstrap to deal with non-normality

of residuals and to get reliable standard errors.

9 In the field of environmental economics, the use of a binary criterion into a meta-analysis was used, among others, by (Jeppesen et al. 2002; Schlpfer 2006; Jacobsen & Hanley 2008; Kiel & Williams 2007; Jordan 2010)

10 It is important to note that, in this paper, we do not look into effects size which would imply using the income associated with turning points.

11 Several authors recommend running OLS when a binary variable is the endogenous regressor. See (Angrist 2001) for more details.

12 Standard errors are clustered by each study. Such approach has been used, for instance, by (Barrio & Loureiro 2010).


15

Table 2. Robust OLS equations

VARIABLES OLS robust OLS cluster-robust Boostrap OLS cluster-

robust

year -0.0220** -0.0220 -0.0220*** (0.00997) (0.0160) (0.00812) initial data 0.00591** 0.00591 0.00591*** (0.00276) (0.00495) (0.00218) end data 0.0111 0.0111 0.0111 (0.00947) (0.0151) (0.00763) sample size 0.000474*** 0.000474*** 0.000474*** (6.13e-05) (9.72e-05) (4.70e-05) H index -0.0686 -0.0686 -0.0686 (0.0799) (0.170) (0.0495) deforestation2 0.483 0.483* 0.483* (0.295) (0.272) (0.249) deforestation3 -0.468*** -0.468** -0.468*** (0.130) (0.184) (0.117) deforestation5 -0.0203 -0.0203 -0.0203 (0.111) (0.158) (0.0909) deforestation6 -0.364** -0.364* -0.364*** (0.149) (0.211) (0.123) deforestation7 -0.188 -0.188 -0.188 (0.154) (0.246) (0.123) deforestation8 -0.468*** -0.468** -0.468*** (0.130) (0.184) (0.119) deforestation9 -0.135 -0.135 -0.135 (0.293) (0.184) (0.334) deforestation10 -0.135 -0.135 -0.135 (0.293) (0.184) (0.340) deforestation11 0.559*** 0.559** 0.559*** (0.175) (0.252) (0.146) deforestation12 -0.0653 -0.0653 -0.0653 (0.162) (0.184) (0.146) deforestation13 -0.410*** -0.410* -0.410*** (0.143) (0.223) (0.127) deforestation14 0.477*** 0.477* 0.477*** (0.162) (0.269) (0.126) econometrics1 0.140 0.140 0.140 (0.159) (0.180) (0.162) econometrics2 -0.684*** -0.684*** -0.684*** (0.165) (0.204) (0.147) econometrics3 0.623*** 0.623** 0.623*** (0.137) (0.255) (0.109) econometrics6 -0.0403 -0.0403 -0.0403 (0.0795) (0.0984) (0.0732) fe_retain 0.0940 0.0940 0.0940 (0.0814) (0.0868) (0.0739) econometrics8 -0.435*** -0.435*** -0.435*** (0.113) (0.143) (0.114) re_retain -0.103 -0.103 -0.103 (0.126) (0.144) (0.114) econometrics10 -0.406*** -0.406*** -0.406*** (0.0920) (0.121) (0.0853) econometrics11 -0.406*** -0.406*** -0.406*** (0.0932) (0.121) (0.0868) econometrics12 0.181 0.181 0.181 (0.257) (0.307) (0.182) area1 0.0487 0.0487 0.0487 (0.0949) (0.146) (0.0859) area2 -0.273** -0.273 -0.273** (0.122) (0.236) (0.124) area3 -0.590*** -0.590*** -0.590*** (0.0724) (0.112) (0.0670) area4 -0.588*** -0.588*** -0.588*** (0.0723) (0.112) (0.0669) area5 0.103* 0.103 0.103** (0.0570) (0.122) (0.0454) area6 -0.00590 -0.00590 -0.00590 (0.240) (0.114) (0.253) area8 -0.0388 -0.0388 -0.0388 (0.0833) (0.136) (0.0694) area10 0.0532 0.0532 0.0532


16

(0.151) (0.159) (0.142) area11 0.135 0.135 0.135 (0.0962) (0.191) (0.0867) area12 -0.117* -0.117 -0.117** (0.0638) (0.0877) (0.0592) development1 0.140** 0.140 0.140*** (0.0656) (0.131) (0.0498) development3 0.661*** 0.661*** 0.661*** (0.140) (0.200) (0.118) development5 0.302* 0.302 0.302* (0.176) (0.186) (0.182) pub type1 0.0232 0.0232 0.0232 (0.0733) (0.138) (0.0528) pub type2 -0.234 -0.234 -0.234 (0.220) (0.244) (0.163) pub type3 -0.250** -0.250 -0.250*** (0.124) (0.191) (0.0946) pub type4 -0.240** -0.240 -0.240*** (0.107) (0.184) (0.0821) life expectancy -0.195* -0.195* -0.195** (0.109) (0.114) (0.0978) Inequality 0.504*** 0.504** 0.504*** (0.150) (0.201) (0.0876) Infrastructures -0.352*** -0.352** -0.352*** (0.106) (0.170) (0.0781) quadratic 0.251** 0.251** 0.251*** (0.106) (0.107) (0.0868) reg 0.000892 0.000892 0.000892 (0.00198) (0.00287) (0.00165) Constant 10.35 10.35 10.35 (11.22) (25.74) (8.122) Observations 631 631 631 R-squared 0.242 0.242 0.242

Robust standard errors in parentheses; *** p


17

Table 3. Variance Inflation Factors

Variables VIF

end data 7.28 infrastructures 2.12 econometrics11 1.33

econometrics6 6.37 re_retain 2.12 econometrics2 1.29

fe_retain 5.96 deforestation12 2.02 deforestation13 1.28

year 5.88 area5 2 area3 1.26

deforestation5 5.42 pub type3 1.7 deforestation10 1.2

pub type4 4.53 development3 1.66 deforestation3 1.2

deforestation7 4.45 area11 1.65 deforestation8 1.2

deforestation2 4.04 econometrics1 1.57 deforestation9 1.2

reg 3.75 econometrics12 1.56 development5 1.13

area12 3.67 inequality 1.54 area6 1.11

deforestation6 3.66 sample size 1.54 deforestation14 1.1

pub type2 3.59 deforestation11 1.5 area2 1.09

pub type1 3.57 area10 1.5 area4 1.09

initial data 3.51 econometrics10 1.49 econometrics3 1.08

H index 3.06 econometrics8 1.46 life expectancy 1.08

development1 2.79 area1 1.39

area8 2.26 quadratic 1.38 Mean VIF 2.44

Secondly, we estimate a Logit model and a Probit model with robust standard

deviations. The choice of one distribution or the other does not induce significant

differences between the results, although tests slightly favour the Probit model.14 Table

4 presents marginal effects for the Probit model.15

14 Tests and Logit estimations are available upon request. See Table A - 2 in the Appendix.

15 Logit and Probit estimations are available upon request to authors. Logit and Probit estimations are run on 546 observations (compared with 613 for OLS) because we have few observations for some variables.


18

Table 4. Probit marginal effects

Robust probit

Probit cluster-robust

Probit Bootstrap cluster-robust VARIABLES

Year -0.0269** -0.0269 -0.0269*** (0.0112) (0.0176) (0.0102) initial data 0.00829*** 0.00829 0.00829*** (0.00316) (0.00531) (0.00274) end data 0.0110 0.0110 0.0110 (0.0104) (0.0164) (0.00879) sample size 0.00209** 0.00209 0.00209*** (0.000812) (0.00173) (0.000711) H index -0.0889 -0.0889 -0.0889 (0.0843) (0.180) (0.0573) deforestation2 0.486** 0.486** 0.486 (0.218) (0.200) (0.507) deforestation5 -0.0805 -0.0805 -0.0805 (0.126) (0.164) (0.114) deforestation7 -0.216* -0.216 -0.216** (0.116) (0.181) (0.105) deforestation9 -0.210 -0.210 -0.210 (0.196) (0.152) (0.137) deforestation10 -0.210 -0.210 -0.210* (0.196) (0.152) (0.127) deforestation12 0.0263 0.0263 0.0263 (0.174) (0.233) (0.155) econometrics1 0.183 0.183 0.183 (0.232) (0.277) (0.181) econometrics6 -0.0771 -0.0771 -0.0771 (0.110) (0.140) (0.108) fe_retain 0.105 0.105 0.105 (0.108) (0.122) (0.114) re_retain -0.134 -0.134 -0.134 (0.135) (0.151) (0.150) econometrics12 0.220 0.220 0.220 (0.279) (0.336) (0.201) area1 0.133 0.133 0.133 (0.143) (0.177) (0.150) area2 -0.264** -0.264 -0.264*** (0.111) (0.167) (0.0793) area5 0.166** 0.166 0.166*** (0.0678) (0.124) (0.0523) area6 0.0312 0.0312 0.0312 (0.233) (0.104) (0.209) area8 0.0259 0.0259 0.0259 (0.136) (0.178) (0.118) area10 0.156 0.156 0.156 (0.175) (0.207) (0.191) area11 0.292** 0.292* 0.292** (0.124) (0.176) (0.118) development1 0.123* 0.123 0.123** (0.0679) (0.133) (0.0573) development5 0.418** 0.418** 0.418*** (0.168) (0.172) (0.159) pub type1 0.0107 0.0107 0.0107 (0.0754) (0.134) (0.0588) pub type2 -0.221 -0.221 -0.221 (0.177) (0.194) (0.499) pub type4 -0.251*** -0.251 -0.251*** (0.0901) (0.153) (0.0802) inequality 0.530*** 0.530*** 0.530*** (0.0950) (0.133) (0.0605) infrastructures -0.297*** -0.297*** -0.297*** (0.0636) (0.106) (0.0514) quadratic 0.245*** 0.245*** 0.245*** (0.0764) (0.0827) (0.0798) reg 0.000182 0.000182 0.000182 (0.00202) (0.00275) (0.00189) Constant Observations 546 546 546 Pseudo R 0.1478 0.1478 0.1478



19

The quality of fit is tested. The quality of the prediction in contingency tables is

reported Table 5. The threshold is fixed at 36 % because in the population (546

observations), there are 36 % of EKC and 64 % of no EKC. Results suggest that we

reach 66.85 % of good predictions with the Probit model.

Table 5. Contingency tables Probit model for EKC

Classified D -D Total + 156 137 293 - 44 209 253

Total 200 346 546 Correctly classified : 66.85%

Classified + if predicted Pr(D) >= .36; True D defined as EKC = 0.

To validate the robustness of our results, we use other thresholds to reject the EKC,

that is to say 1% and 5% instead of 10% retained in this work. In other words, we build

two new dependent variables EKC01 and EKC05. With the first one, we choose a

significance level of 1% for the variables GDP per capita and its square. With the second

we use a threshold of 5%. Using the 10 % threshold we have 32 % of the observations

validating the EKC. It falls to 27 % for the 5 % threshold and 17 % for the 1 % one. We

then estimate the Probit model. Results do not significantly change.16

6. Results and Discussion

6.1. Main results

Our results show that the most recent year of publication is (year), the more likely to

have an EKC declines. We shall then identify the existence of non-linearity with respect

to the year of publication. In the OLS regressions (robust, cluster and bootstrap cluster),

we add the square of the year of publication. Both variables year and its square appear

to be significant in the three models, the first one having a positive coefficient and the

second having a negative one. In addition, coefficients are the same in each model

16 Tables are available from the authors upon request (See Table A 3 in the Appendix)


20

(respectively 18.89 and 0.00472). The turning point is around the year 2001.17,18 This

may be attributable to several factors, including an academic fad and technology

(econometric practices and better availability of data).

Empirical choices should be carefully done before deriving policies from empirical

results. On this issue, our results are not clear-cut which is not in line with (D. I. Stern

2004) contention according to which early empirical evidences of EKCs where the result

of weak econometrics.19 It is clear however from OLS regressions that the integration of

a spatial dimension (econometrics10 econometrics11) decreases the likelihood of an

EKC.20 Although these results are not to be taken per se, they raise the question of the

existence of spillovers between the rates of deforestation of countries. If these spillovers

exist, does it change the nature of the relationship between economic development and

deforestation? Existing work seem to go in that direction (e.g. Nguyen Van & Azomahou

2007).

We therefore believe that future work should incorporate the spatial dimension in

order to control for the existence of spatial autocorrelation (Anselin 1988). Regarding

deforestation, there may be several sources of spatial dependence between countries.

First, national policies for forest conservation may be influenced by policies in

neighboring countries resulting in a pattern of political spatial dependence. Second,

there may be positive or negative ecological spillovers between countries. For instance, 17 18.89/(2*0.00472) = 2001.06

18 Our robustness check provides consistent turning points. We precisely find 1999.29 for 1 %, 2000.98 for 5 % and 2000.39 for 10 %.

19 The existence of an EKC for deforestation was seldom performed on time series data. It was however the case with other types of pollutants such like CO2 or SO2. Even with recent econometric techniques, the results on the existence of an EKC are not conclusive (see e.g. Bernard et al. 2011).

20 It is likely that the number of spatial regressions is too small to be captured in Logit and Probit estimations. We run a regression with three dummies for econometric techniques: OLS, panel (dynamic, fixed effects and random effects) and spatial (panel and GWR). It did not change the results. OLS and panel estimations were not significant, whereas spatial estimations were negative and significant.


21

seeds or forest fires if necessary to recall do not respect political boundaries. Third,

unobserved variables may be related by a spatial process. In the case of deforestation,

these may be climatic or geomorphological variables. And fourth, phenomena in

neighboring countries are likely to influence deforestation patterns in a given country.

For example, more stringent forest protection policies in a country may cause more

deforestation in an adjacent country. As a matter of fact, there is uncertainty about the

choice of the relevant spatial model (i.e. whether to estimate a spatial lag model, a

spatial error model, a spatial Durbin model or a Cliff-Ord type model). All effects can

occur simultaneously and there is no theoretical argument to exclude a form of spatial

dependence a priori. The use of spatial econometrics is relatively recent. Theoretical

(Corrado & Fingleton 2011; Elhorst 2010) and technical literature is growing fast

(Drukker et al. 2011). Regarding the EKC for deforestation thanks to the better

availability of data, their use will be facilitated and more relevant. This is an avenue of

research to dig that could call into question the existence of an EKC for deforestation.

At that point, we cannot conclude on the quality of data although this concern was put

forward (D. I. Stern et al. 1996).21 Still, we do find that the more recent data coincide

with the more chance of getting an EKC. It therefore stems the following question: Is it

appropriate to mix data of different quality? Our analysis cannot answer that question.

6.2. Five remarks

Sample size

The sample size (sample size) has a positive and significant effect on the existence of

the EKC. This result is line with other meta-analyses (Jeppesen et al. 2002). Indeed,

21 When gathering information we have extracted the data source. However, this information could not be used in the meta-analysis because these variables are highly correlated with measures of deforestation and because their source very often relies on FAO data.


22

according to standard testing theory sample sizes increase the power of tests. In other

words, larger sample sizes reduce the probability of erroneously accepting the null

hypothesis.

Deforestation index

The term deforestation is not clear-cut. It covers a wide range of meanings. This is

striking in EKC studies for deforestation. Authors use different measures (Table 1). Our

results do not evidence a clear effect of the deforestation index on the probability of the

occurrence of an EKC for deforestation.

Infrastructures

Deforestation is driven by numerous factors. In some studies, authors only focus on

the income-deforestation nexus. In others, authors add control variables. We find that

adding infrastructures variables (infrastructures) decreases the probability of obtaining

an EKC. There is evidence that infrastructure expansion is a cause of deforestation (Pfaff

1999). However, this variable may be highly correlated to income. (D. I. Stern 2004)

addresses this issue and concludes that it is difficult to infer from studies that include

additional control variables. We add that our results for infrastructures are not

surprising. Indeed, if they are correlated with per capita income, it is normal that

authors do not find a significant relationship between economic development and

deforestation. Yet, we argue that it is not enough to conclude that there is no EKC in that

specific case, because such result could also be explained by the fact that this variable

plays a role in channelling the effect of economic development on deforestation.


23

Inequality index

We find a strong positive relationship between the introduction of a control variable

for inequalities (inequality) and the probability of finding an EKC. Our result does not

contradict (Koop & Tole 2001) and (Heerink et al. 2001). They find that high inequality

strengthens the effect of economic development on deforestation.

Publication type

If the type of publication is a Master's thesis (pub type4), then the probability of an

EKC declines relative to an article, a working paper or conference paper, ceteris paribus.

This could be explained by a publication strategy effect. A Masters student seems not

to have the same incentives to validate the existence of an EKC whereas a researcher

may be more prone to evidence an EKC as a means to publish his work in an academic

journal.

6.3. Publication Bias

Publication bias remains today the biggest problem facing meta-analyses. Indeed in

principle, the meta-analysis must combine exhaustively all the studies that have been

made in the area of interest. First there may be a bias in the reporting of results: the

estimation results are more likely to be published if they are significant. This problem is

referred as the file drawer problem by meta-analysts (Stanley et al. 2008). Non-

significant results may not be submitted to academic journals (nor published as a

working paper) or they may be more rejected by publishers of academic journals. Thus

the literature is not a perfect representation of all work in the field. The sources of this

type of bias are complex and highly dependent on the decisions of the authors as well as


24

publishers. Secondly, studies with good results are often published in English. Here,

the means of dissemination may also be a source of publication bias.

Also, in our analysis we do not take into account the quality of the papers. This

problem is referred to as garbage in, garbage out (Borenstein et al. 2009). However,

we believe that excluding articles, on its own, would only add bias, compared to what we

believe to be a good paper. In addition, our work allows rather performing waste

management.22 In this sense, our work examines the characteristics of the studies that

are related to the occurrence of results.

7. Concluding Remarks

The objective of this paper was to shed light on why EKC for deforestation results

differ across studies. And that, in view of raising the attention of researchers,

practitioners and policy makers on the underlying causes that lead to the validation of

the EKC for deforestation in the academic literature, despite critical reviews of EKCs. We

have therefore constructed an original database gathering the characteristics of

econometric results for EKCs for deforestation and then implemented a meta-analysis.

Our main results are the following. Recent estimates do not corroborate the EKCs for

deforestation. The year 2001 appears to be a turning point. Our results also confirm that

empirical choices i.e. estimators choice affect the occurrence of an EKC for deforestation

with respect to the hypothesis of spatial autocorrelation in deforestation rates. The

contention of (D. I. Stern 2004) of the poor quality of econometrics underlying EKCs

estimates is thus given a particular flavour.

22 In the metaphorical sense proposed by (Borenstein et al. 2009)


25

Finally, let us consider another interpretation of our results in a Popperian way. A

reason why EKCs are still considered is that scholars hardly make a definitive refutation

of the EKC hypothesis; they rather prefer evidencing the presence or the absence of

corrobation. Indeed, in order to test the relevance of the theory underlying the EKC,

scholars must consider auxiliary hypotheses. In the case of the EKC those auxiliary

hypotheses are partly provided by the econometric hypotheses. When the EKC is not

corroborated, scholars cannot identify whether it is the result of falsified theory or of the

econometric hypotheses. This is the essence of Quines holism (Quine 1951): the EKC for

deforestation cannot be tested in isolation.23 The EKC story is not at its end. Building on

our results, we can predict that empirical and theoretical research on EKCs is still lively

until theoretical alternatives will be provided. Following Popper, we can hope that

refuted theory can be repaired to cope with the newly discovered anomalies (Blaug

1992, p.25)

Lastly, as mentioned earlier, a meta-analysis opens the way for prospective meta-

analysis. In future work, we shall explore the quality of estimations. Indeed, in this

article we focused on estimators. Notably, we shall focus on EKC studies using panel

techniques and investigate the study-to-study variation focusing on econometric tests

performed such as tests for endogeneity, multicollinearity, Hausman test, etc.

23 The critic of Quine is also referred as the Duhem-Quine critic on the confirmation holism.


26

References Andreoni, J. & Levinson, A., 2001. The simple analytics of the environmental Kuznets curve. Journal of

Public Economics, 80(2), p.269286.

Angelsen, A., Brown, S., et al., 2009. Reducing Emissions from Deforestation and Forest Degradation (REDD): an options assessment report. Prepared for the Government of Norway, Meridian Institute.

Angelsen, A., Brockhaus, M., et al. d., 2009. Realising REDD+. National Strategy and Policy Options, Bogor: Center for International Forestry Research.

Angelsen, A. & Kaimowitz, D., 1999. Rethinking the Causes of Deforestation: Lessons from Economic Models. The World Bank Research Observer, 14(1), p.7398.

Angrist, J.D., 2001. Estimation of Limited Dependent Variable Models with Dummy Endogenous Regressors: Simple Strategies for Empirical Practice. Journal of Business & Economic Statistics, 19(1), p.216.

Anselin, L., 1988. Spatial econometrics: methods and models, Dordrecht: Kluwer Academic Publishers.

Arrow, K. et al., 1996. Economic Growth, Carrying Capacity, and the Environment. Environment and Development Economics, 1(01), p.104110.

Barbier, E.B., 2001. The Economics of Tropical Deforestation and Land Use: An Introduction to the Special Issue. Land Economics, 77(2), p.155171.

Barbier, E.B., Burgess, J.C. & Grainger, A., 2010. The forest transition: Towards a more comprehensive theoretical framework. Land Use Policy, 27(2), p.98107.

Barrio, M. & Loureiro, M.L., 2010. A meta-analysis of contingent valuation forest studies. Ecological Economics, 69(5), p.10231030.

Beckerman, W., 1992. Economic growth and the environment: Whose growth? whose environment? World Development, 20(4), p.481496.

Bernard, J.-T. et al., 2011. The Environmental Kuznets Curve: Tipping Points, Uncertainty and Weak Identification. SSRN eLibrary. Available at: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1974622.

Blaug, M., 1992. The Methodology of Economics. Or, How Economists Explain 2nd ed., Cambridge University Press.

Borenstein, M. et al., 2009. Introduction to Meta-Analysis, Chichester UK: John Wiley and Sons.

Carson, R.T., 2010. The Environmental Kuznets Curve: Seeking Empirical Regularity and Theoretical Structure. Review of Environmental Economics and Policy, 4(1), p.3 23.

Cavlovic, T.A. et al., 2000. A meta-analysis of environmental Kuznets curve studies. Agricultural and Resource Economics Review, 29(1), p.3242.

Chatterjee, S. & Hadi, A.S., 2006. Regression Analysis by Example 4th ed., John Wiley and Sons.

Corrado, L. & Fingleton, B., 2011. Where Is the Economics in Spatial Econometrics? Journal of Regional Science, p.130.

Dinda, S., 2004. Environmental Kuznets Curve Hypothesis: A Survey. Ecological Economics, 49(4), p.431455.

Drukker, D.M., Prucha, I.R. & Raciborski, R., 2011. A command for estimating spatial-autoregressive models with spatial-autoregressive disturbances and additional endogenous variables. The Stata Journal, 1(1), p.113.

Elhorst, J.P., 2010. Applied Spatial Econometrics: Raising the Bar. Spatial Economic Analysis, 5(1), p.928.

Engel, S. & Palmer, C., 2008. Painting the forest REDD? Prospects for mitigating climate change through reducing emissions from deforestation and degradation. IED Working paper, (3).

Geist, H.J. & Lambin, E.F., 2001. What drives tropical deforestation? A meta-analysis of proximate and underlying causes of deforestation based on subnational case study evidence, Louvain-la-Neuve: LUCC International Project Office - University of Louvain.


27

Grossman, G.M. & Krueger, A.B., 1995. Economic Growth and the Environment. The Quarterly Journal of Economics, 110(2), p.353377.

Heerink, N., Mulatu, A. & Bulte, E., 2001. Income inequality and the environment: aggregation bias in environmental Kuznets curves. Ecological Economics, 38(3), p.359367.

Hunter, J.E. & Schmidt, F.L., 2004. Methods of Meta-Analysis. Correcting Errors and Bias in Research Findings, Sage.

Jacobsen, J.B. & Hanley, N., 2008. Are There Income Effects on Global Willingness to Pay for Biodiversity Conservation? Environmental and Resource Economics, 43(2), p.137160.

Jeppesen, T., List, J.A. & Folmer, H., 2002. Environmental Regulations and New Plant Location Decisions: Evidence from a MetaAnalysis. Journal of Regional Science, 42(1), p.1949.

Jordan, B.R., 2010. The Environmental Kuznets Curve: Preliminary Meta-Analysis of Published Studies, 1995-2010. Dans Workshop on Original Policy Research. Georgia Tech - School of Public Policy.

Kiel, K.A. & Williams, M., 2007. The impact of Superfund sites on local property values: Are all sites the same? Journal of Urban Economics, 61(1), p.170192.

Koop, G. & Tole, L., 2001. Deforestation, distribution and development. Global Environmental Change, 11(3), p.193202.

Koop, G. & Tole, L., 1999. Is there an environmental Kuznets curve for deforestation? Journal of Development Economics, 58(1), p.231244.

Kuhn, T.S., 1996. The structure of scientific revolutions 3rd ed., Chicago: University of Chicago Press.

Levinson, A., 2002. The ups and downs of the environmental Kuznets curve. In J. A. List & A. de Zeeuw, ed. Recent advances in environmental economics. London: Edward Elgar, p. 119.

Lindhjem, H., 2007. 20 years of stated preference valuation of non-timber benefits from Fennoscandian forests: A meta-analysis. Journal of Forest Economics, 12(4), p.251277.

Ma, C. & Stern, D.I., 2006. Environmental and ecological economics: A citation analysis. Ecological Economics, 58(3), p.491506.

Mariani, F., Prez-Barahona, A. & Raffin, N., 2010. Life expectancy and the environment. Journal of Economic Dynamics and Control, 34(4), p.798815.

Mather, A.S., 1992. The Forest Transition. Area, 24(4), p.367379.

Mather, A.S., Needle, C.L. & Fairbairn, J., 1999. Environmental Kuznets Curves and Forest Trends. Geography, 84(1), p.5565.

McCloskey, D.N., 1983. The Rhetoric of Economics. Journal of Economic Literature, 21(2), p.481517.

Munasinghe, M., 1999. Is environmental degradation an inevitable consequence of economic growth: tunneling through the environmental Kuznets curve. Ecological Economics, 29(1), p.89109.

Nelson, J.P. & Kennedy, P.E., 2008. The Use (and Abuse) of Meta-Analysis in Environmental and Natural Resource Economics: An Assessment. Environmental and Resource Economics, 42(3), p.345377.

Nguyen Van, P. & Azomahou, T., 2007. Nonlinearities and heterogeneity in environmental quality: An empirical analysis of deforestation. Journal of Development Economics, 84(1), p.291309.

Panayotou, T., 1993. Empirical Tests and Policy Analysis of environmental Degradation at Different Stages of Economic Development, Geneva: International Labor Office.

Pfaff, A.S.P., 1999. What Drives Deforestation in the Brazilian Amazon?: Evidence from Satellite and Socioeconomic Data. Journal of Environmental Economics and Management, 37(1), p.2643.

Quine, W.V.O., 1951. Main Trends in Recent Philosophy: Two Dogmas of Empiricism. The Philosophical Review, 60(1), p.2043.

Roy Chowdhury, R. & Moran, E.F., 2012. Turning the curve: A critical review of Kuznets approaches. Applied Geography, 32(1), p.311.

Rudel, T.K. et al., 2005. Forest transitions: towards a global understanding of land use change. Global Environmental Change, 15(1), p.2331.


28

Rudel, T.K. & Roper, J., 1997. The paths to rain forest destruction: Crossnational patterns of tropical deforestation, 1975-1990. World Development, 25(1), p.5365.

Schlpfer, F., 2006. Survey protocol and income effects in the contingent valuation of public goods: A meta-analysis. Ecological Economics, 57(3), p.415429.

Shafik, N. & Bandyopadhyay, S., 1992. Economic Growth and Environmental Quality. Time-Series and Cross-Country Evidence. Policy Research Working Papers Series, 904.

Stanley, T.D., 2001. Wheat from Chaff: Meta-Analysis as Quantitative Literature Review. The Journal of Economic Perspectives, 15(3), p.131150.

Stanley, T.D., Doucouliagos, C. & Jarrell, S.B., 2008. Meta-regression analysis as the socio-economics of economics research. Journal of Socio-Economics, 37(1), p.276292.

Stern, D.I., 2004. The Rise and Fall of the Environmental Kuznets Curve. World Development, 32(8), p.14191439.

Stern, D.I., Common, M.S. & Barbier, E.B., 1996. Economic growth and environmental degradation: The environmental Kuznets curve and sustainable development. World Development, 24(7), p.11511160.

Stern, N.H., 2007. The economics of climate change: the Stern review, Cambridge University Press.

Xepapadeas, A., 2005. Chapter 23 Economic growth and the environment. Dans K.-G. Mler & J. R. Vincent, d. Economywide and International Environmental Issues. Elsevier, p. 12191271.


29

Appendix A For the referees, not for publication

Table A 1. Studies included in the sample (authors, year of publication, support of the publication)

Antle & Heidebrink (1995) Economic Development and

Cultural Change

Araujo et al (2009) Ecological Economics

Arcand et al (2008) Journal of Development Economics

Arcand et al (2002) CERDI working papers

Barbeau (2003) University de Montreal, Dpartment de

sciences conomiques, Rapport de recherche

Barbier (2004) American Journal of Agricultural

Economics

Barbier & Burgess (2001) Journal of Economic Surveys

Basu & Nayak (2011) Forest Policy and Economics

Bhattarai & Hammig (2004) Environment and

Development Economics

Bhattarai & Hammig (2001) World Development

Castro Gomes & Jos Braga (2008) Universidade

Federal De Viosa-UFV, Working papers

Buitenzorgy & Mol (2011) Environmental and Resource

Economics

Caviglia-Harris et al (2009) Ecological Economics

Combes et al (2008) CERDI working papers

Combes et al (2010) CERDI working papers

Combes Motel et al (2009) Ecological Economics

Corderi (2008) Economic Analysis Working Papers,

Colexio de Economistas de A Corua, Spain and

Fundacin Una Galicia Moderna

Cornelis van Kooten & Wang (2003) XII World Forestry

Congress (FAO)

Corra de Oliveira & Simoes de Almeida (2011)

Department of Economics- Federal University of Juiz

Juiz de Fora

Cropper & Griffiths (1994) American Economic Review

Culas & Dutta (2003) Working papers, University of

Sydney, Department of economics

Culas (2007) Ecological Economics

Damette & Delacote (2011) Ecological Economics

Delacote (2003) Mmoire de DEA CERDI

Diarra & Marchand (2011) CERDI working papers

Diarrasouba & Boubacar (2009) Selected paper for

presentation at the southern Agricultural economics

Association Annual Meeting

Duval & Wolff (2009) Laboratoire d'Economie et de

Management Nantes-Atlantique Universit de Nantes,

HAL working Papers

Ehrhardt-Martinez et al (2002) Social Science Quarterly

Ehrhardt-Martinez (1998) Social Forces

Ferreira (2004) Land Economics

Frankel & Rose (2002) NBER Working papers

Galinato & Galinato (2010) Working papers, School of

Economic Sciences, Washington State University

Gangadharan & Valenzuela (2001) Ecological

Economics

Heerink et al (2001) Ecological Economics

Hilali & Ben Zina (2007) Unit de Recherches sur la

Dynamique Economique et l'Environnement (URDEE)

Hyde et al (2008) Environment for Development

Jorgenson (2006) Rural Sociology

Jorgenson (2008) The Sociological Quarterly

Kahuthu (2006) Environment, Development and

Sustainability

Kallbekken (2000) Environmental Economics and

Environmental Management, University of York

Kishor & Belle (2004) Journal of Sustainable Forestry

Koop & Tole (1999) Journal of Development Economics

Lantz (2002) Journal of Forest Economics

Lee et al (2009) Singapore Economic review conference

Leplay & Thoyer (2011) Laboratoire d'Economie et de

Management Nantes-Atlantique Universit de Nantes,

HAL working Papers

Li (2006) International Studies Quarterly

Lim (1997) School of economics, University of New

South Wales Sydney, NSW 2no52, Australia

Lopez & Galinato (2004) Environmental Protection

Agency (EPA) report

Marchand (2010) CERDI working papers

Marquart-Pyatt (2004) International Journal of Sociology

Meyer et al (2003) International Forestry Review

Naito & Traesupap (2006) Journal of The faculty of

Economics, KGU

Nguyen Van & Azomahou (2003) Revue Economique

Nguyen Van & Azomahou (2007) Journal of

Development Economics

Panayotou (1993) World Employment Programme

research Working papers

Perrings & Ansuategi (2000) Journal of Economic

Studies

Reuveny et al (2010) Journal of Peace Research

Rock (1996) Ecological Economics

Shafik (1994) Oxford Economic Papers

Shafik & Bandyopadhyay (1992) Policy Research

Working papers, World Bank

Shandra (2007) Sociological Inquiry

Shandra (2007) International Journal of Comparative

Sociology

Shandra (2007) Social Science Quarterly

Skonhoft & Solem (2001) Ecological Economics

Turner et al (2006) Scandinavian Journal of Forest

Research

Turner et al (2005) Agricultural and Resource Economics

Review

Wang et al (2007) Forest Policy and Economics

Wang (2003) XII World Forestry Congress 2nono3

(FAO)

Yoshioka (2010) Faculty of economics, Nagasaki

University

Zhao et al (2011) Environment and Resource Economics

Zhu & Zhang (2006) Journal of Agricultural and Applied

Economics


30

Table A 2. Logit Marginal Effects

Robust logit VARIABLES

Year -0.0276** (0.0123) initial data 0.00818** (0.00339) end data 0.0113 (0.0110) sample size_section 0.00220** (0.000888) H index -0.0801 (0.0865) deforestation2 0.491** (0.212) deforestation5 -0.0897 (0.137) deforestation7 -0.209* (0.113) deforestation9 -0.200 (0.179) deforestation10 -0.200 (0.179) deforestation12 0.0257 (0.181) econometrics1 0.178 (0.242) econometrics6 -0.0886 (0.116) fe_retain 0.0869 (0.114) re_retain -0.154 (0.134) econometrics12 0.188 (0.308) area1 0.151 (0.157) area2 -0.248** (0.117) area5 0.178** (0.0720) area6 0.0330 (0.233) area8 0.0299 (0.148) area10 0.173 (0.188) area11 0.319** (0.134) development1 0.124* (0.0705) development5 0.432*** (0.151) pub type1 0.00568 (0.0766) pub type2 -0.238 (0.163) pub type4 -0.271*** (0.0860) inequality 0.523*** (0.0954) infrastructures -0.279*** (0.0649) quadratic 0.237*** (0.0769) reg 0.000404 (0.00211)

Observations 546 Pseudo R 0.1472



31

Table A - 3. Probit robustness check VARIABLES EKC01 EKC05

Year -0.0936*** -0.0630** (0.0361) (0.0320)

initial data 0.0378*** 0.0195** (0.0118) (0.00865)

end data 0.0104 0.0111 (0.0382) (0.0321)

sample size 0.00556* 0.00472** (0.00319) (0.00211)

H index -0.175 -0.0494 (0.301) (0.254)

deforestation2 4.822*** 0.710 (0.736) (0.934)

deforestation5 -1.314*** -0.886** (0.405) (0.355)

deforestation7 -2.194*** -1.577*** (0.698) (0.486)

deforestation9 -1.019 -0.933 (0.830) (0.811)

deforestation10 -1.019 -0.933 (0.830) (0.811)

deforestation11 -1.009 0.686 (0.925) (0.898)

deforestation12 -0.699 (0.576)

econometrics1 0.249 0.483 (0.661) (0.613)

econometrics6 -0.559 0.0765 (0.402) (0.322)

fe_retain -0.125 0.278

(0.366) (0.314)

re_retain -0.798 -0.497

(0.674) (0.500)

econometrics12 1.743** 1.101 (0.760) (0.737)

area1 0.261 0.225

(0.486) (0.375)

area2 -0.894 -0.916

(0.655) (0.587)

area5 0.469** 0.456***

(0.208) (0.175)

area6 -0.385

(0.606)

area8 0.168 -0.151

(0.479) (0.411)

area10 0.0680

(0.451)

area11 -0.0783 0.565*

(0.490) (0.334)

development1 0.852*** 0.389** (0.224) (0.195)

development5 0.922 1.356** (0.683) (0.605)

pub type1 0.269 0.0305 (0.269) (0.215)

pub type2 -4.576*** -0.273 (0.355) (0.768)

pub type4 -1.598*** -1.170*** (0.487) (0.386)

inequality 1.527*** 0.789*

(0.432) (0.428)

infrastructures -12.35 -5.542***

(8.348) (1.912)

quadratic 0.771** 0.828** (0.392) (0.347)

reg 0.0169** 0.0101* (0.00686) (0.00600)

Constant 90.88** 64.09**

(36.65) (31.70)

Observations 512 550

Pseudo R 0.2908 0.1873 Robust standard errors in parentheses; *** p

The Environmental Kuznet Curve for Deforestation

Documents

Transcript of The Environmental Kuznet Curve for Deforestation